Predicting Violence Against Women Using the Women’s Economic Opportunity Index (WEOI)

An exploration of the the main indicators of the WEOI and what they can tell us about violence against women

Introduction

Even though “achieving gender equality and empowering all women and girls” is among one of the 17 Sustainable Development Goals by the United Nations (“Goal 5 Department of Economic and Social Affairs” n.d.), such goal swims against an enormous wave of violence against women in the contemporary world. Investigation on the causes of such violence is neither exhaustive nor simple. This paper aims to investigate what are possible reasons for the bugging persistence of violence against women, specifically in what regard to a positive attitudes towards violence against women, i.e., more women finding justifiable that a husband would beat their wives.

Believing in the importance of promoting Women’s economic well-being in order to achieve equality, we aim to understand the relationship between violence against women and the Women’s Economic Opportunity Index (WEOI). Providing a violence-free environment for women and girls will immensely support their empowerment, which in turn represents countless gains for society as a whole (King and Mason 2001).

After a brief literature review on the subject of women’s economic well-being, we explain our two-part methodology. We pulled our data from the Organisation for Economic Co-operation and Development (OECD), the United Nations (UN), the World Bank (WB), and the Economist Intelligence Unit (EIU). Our analysis reveals that even though women’s legal status seems to be one of the best predictors within the WEOI for attitude towards violence against women, we must be attentive to the ways in which indicators interact within and outside of the Women’s Economic Opportunity Index affect each other.

Literature Review

Women are disadvantaged in entrepreneurship. Part of that is because structures that push for entrepreneurship have for long pretended to be “genderless” when they are not (Pathak, Goltz, and W. Buche 2013). It is well-known that organizations, by not identifying or addressing the barriers that women might uniquely face, such as child-rearing and domestic labour (Acker 1990), have failed to provide equality in the workplace. Our research dialogues with previous literature in asserting that difficulty in women’s access to finance and labor participation further limit the possibility of women gaining independence from, at best, their domestic and child-rearing duties, and at worse, from their partners. By constraining women from participating in labor in equal and fair manners, as well accessing finance, we argue that this could be increasing violence against women.

Women’s education and legal status can both potentially decrease levels of violence against women. When education is intentionally designed to address gender inequality, it has the potential to provide the tools women need to both identify and report experiences of violence (Okenwa-Emgwa and Strauss 2018). Similarly, women’s legal status is especially important because it provides them with access to institutions and legal frameworks they can resort to when experiencing violence (Tavares and Quentin 2018). Not only does “legal status” include explicit regulation against violence experienced by intimate partners, but also legal frameworks protecting their rights to freedom of movement, property ownership, adolescent fertility, and regulations addressing all forms of discrimination (“Women’s Economic Opportunity 2012” n.d.). However, the literature has raised questions on the effectiveness of both education and legal status in raising the stature of women and preventing violence alone. For example, even though the benefits of education alone are evident (“Girl Rising Girls Education Nonprofit” n.d.), in order to effectively reduce violence against women, it is necessary to formulate curricula that include gender equality and information on violence against women (Okenwa-Emgwa and Strauss 2018). In the case of women’s legal status, adequate implementation and observation of such laws are needed for them to be effectively enforced (Tavares and Quentin 2018). Our research interacts with such limitations by exploring whether WEOI’s indicators on education and training and women’s legal status can predict a decrease in positive attitude towards violence.

Methodology

This study consists of two parts. The first part is an exploration of three indicators that the Organisation for Economic Co-operation and Development (OECD) has linked to Violence Against Women. The first indicator is Attitude Towards Violence, which represents the share of women who agree that a husband is justified in beating his wife. The second is Prevalence of Violence in Lifetime (0-100%), which indicates the share of women exposed to physical and/or sexual violence from an intimate partner at least once in their lives. Finally, Laws on Domestic Violence (0.25-1) measures whether legal frameworks adequately protect women from domestic violence. It is measured on a scale from 0 to 1, where 1 indicates that the laws are completely discriminatory against women.

The second part of this research uses the Women’s Economic Opportunity Index (WEOI) to predict violence against women. Created by the Economist Intelligence Unit, the research division of the Economist Group, this index “is a dynamic quantitative and qualitative scoring model, constructed from 29 indicators, that measures specific attributes of the environment for women employees and entrepreneurs in 128 economies” (“Women’s Economic Opportunity 2012” n.d.). The index is measured on a scale from 0 - 100 where 0 refers to the least favorable situation for women and 100 refers to the most favorable situation to women.

We dissect the index into four out of its five main variables (labour participation, access to finance, women’s legal status, and women’s education and training), and we use these four factors as our independent variables. Our dependent variable is attitudes towards violence against women. We draw a causal graph to predict the factors involved in causing negative attitudes towards violence, and then we then run several univariate and multivariate regression models were between our dependent and independent variables.

We conclude that from our four main indicators, Women’s Legal Status seems to have the biggest potential to predict a decrease in attitude towards violence. From there, we highlight subvariables within women’s legal status that have a relationship with our main dependent variable. Finally, we analyze the causal relationship between the democracy index and the legal status of women, which may be influencing attitudes towards women

Theory/Hypotheses

With the literature in mind, and with a curiosity towards measurements taken by the Women’s Economic Opportunity Index, our research question consists on how do labour participation, women’s legal status, access to finance and women’s education and training appear to reduce or increase violence against women? Specifically, which one of those four variables seem to be the best predictor for an increase or reduction in attitude towards violence?

Even though we explore four out of the five indicators used in WEOI in order to possibly find out which indicators seem to best predict attitude towards violence, we also pay attention to the interactions between such indicators. We do so because we are well aware of possible confounding variables. We have decided to drop the fifth main indicator “General Business Environment.” Whereas this could be a good indicator for observing practical obstacles for the women in the workplace, we thought that focusing on the first four indicators could make our research more consistent.

Labour participation consists in four indicators: equal pay for equal work, non-discrimination, degree of de facto discrimination against women in the workplace, and availability, affordability, and, quality of childcare services (“Women’s Economic Opportunity 2012” n.d.)

Access to finance consists in four indicators: building credit histories, women’s access to finance programmes, delivering financial services, private-sector credit as a percent of Gross Domestic Product (“Women’s Economic Opportunity 2012” n.d.).

Education and training consists of four indicators: School life expectancy (primary and secondary), school life expectancy (tertiary), adult literacy rate, and existence of government or non-government programmes offering small and medium-sized enterprise (SME) support/development training (“Women’s Economic Opportunity 2012” n.d.).

Women’s legal status consists of five indicators: existence of laws protecting women against violence, freedom of movement, property ownership rights, adolescent fertility rate, and whether a country ratification of the Convention on the Elimination of All Forms of Discrimination against Women (CEDAW) (“Women’s Economic Opportunity 2012” n.d.).

Data

Part 1: Exploration of Dependent Variables

Visualizing Dependent Variables: The OECD Violence Against Women Indicator

In our first stage in our data exploration, we decide to observe whether there were patterns in a country’s economy and our dependent variables. We expected that we would find patterns because a more developed economy might mean more economic opportunities for women.

There seems to be a clear association between attitude towards violence against women and a country’s income group. High and upper middle income countries seem to have a lower share of women who agree that the husband is justified in inflicted violence in their wives.

Similar to Attitudes Towards Violence, when it comes to prevalence of violence in the lifetime, we can observe that high income countries (colored red) are concentrated towards the left side of the graph. This means that fewer women have been exposed to physical or sexual violence in comparison to countries falling in other income groups.

Whereas in regards to the previous variables it was clearer to see the distribution of data in regards to income group, Laws on domestic violence doesn’t seem to follow the same pattern. The graph shows that most countries, regardless of income group, fall in the “0.75” measure for Laws on Domestic Violence. That means that their laws and practices are unsuficcient to garantee the well-being of women. It is evident that even though there are few countries in which the number reaches the value 1 (laws and practices are completely discriminatory against women), the realization that most countries in the globe still have unsuficcient laws that protect women’s rights is incredibly worrying. We can also observe that there are more countries with better laws and practices for women’s rights among high income countries (that is, countries in which the indicator equals 0.25). This also holds true in a lesser extent for upper middle income countries. For both lower middle income and low income, however, there seems to be more countries in the 0.5 and 0.75 levels then in the 0.25.

This unveils a pattern that establishes that higher income coutries, despite also having large number of countries that have a indicator of law on domestic violence equal to 0.75, have also higher number of countries that have a somewhat benefitial laws and practices against domestic violence. However, this does not seem to be a causal effect. It is highly unlikely that a countrie’s income group alone establishes the laws on domestic violence, but it points us in the direction of establishing a connection between higher levels of economic develpment and better legislation on domestic violence.

We proceed to design stacked bars in order to better understand patterns in the data, specially when it came to laws on domestic violence. However, one important aspect to notice is that the total number of countries in each income category will affect the visualization of these stack bars when they are not designed to take into account proportion. See below:

# A tibble: 4 × 2
  IncomeGroup         count
  <fct>               <int>
1 High income            39
2 Upper middle income    33
3 Lower middle income    37
4 Low income             19

There are 39 high income, 33 upper middle income, 37 lower middle income, and only 19 lower income countries in the data set.

A stacked bar helps in the visualization and reveals that indeed most of the countries that have a value of 0.25 on Laws on Domestic Violence (better laws on domestic violence) are high income countries. However, it is important to notice that in the 0.75 value, there seems to be a more balanced distribution throughout all income groups. In the value of “1,” i.e., laws on domestic violence that completely discriminate against women, we find only upper middle income countries.

We also produced a bar stack for prevalence of violence in the lifetime, but instead of using a proportion stacked bar we opted for a frequency one. Even though this plot suffers from the total number of high income countries being superior overall, the concentration of such countries in the left side of the graph once again reveal the pattern of high income countries performing better in such indicators.

After drawing analysis by Income Group, we make a comparison between all three indicators to determine their relationship to one another.

[1] 0.4489546

X-axis shows Prevalence of Violence in Lifetime, Y-axis shows Attitude Towards Violence, and the colored dots indicate Laws Against Domestic Violence. The visualization shows a moderate positive relationship between Prevalence of Violence in the Lifetime and Attitude Towards Violence (correlation coefficient = 0.45). We can also visualize that yellow-dots countries are concentrated in the left lower side of the plot, thus revealing that countries that have decent laws on Domestic Violence also tend to perform better in the other two indicators.

The annotated countries in the graph are countries in which the Laws Against Domestic Violence are completely discriminatory against women: Equatorial Guinea and the Russian Federation.Equatorial Guinea, besides having very discriminatory laws against domestic Violence, also has higher levels of Prevalence of Violence and Attitude Towards Violence. On the other hand, the Russian Federation is an interesting case. It shows lower levels of Prevalence of Violence and Attitude Towards Violence, but its legislation seem to completely discriminate against women. It is closer to the yellow dots on the graph, which are countries with good Laws on Domestic Violence. This sets Russia apart as an notable outlier when plotting the relationship between the three dependent variable. Research on its case will be interesting when analyzing potential causes for the persistence of violence against women.

Comparing OECD and UN in Prevalence of Violence Againt Women

To determine whether the OECD indicator is an accurate representation of the prevalence of domestic violence against women, we decided to compare it to other available indicators on domestic violence. We have chosen the UN indicator, “Proportion of women subjected to physical and/or sexual violence by a current or former intimate partner in the last 12 months.”

There are some clear distinctions between the UN and OECD indicators. While they both measure the proportion of women who have experienced violence by an intimate partner, the UN indicator only includes women who have experienced violence in the past 12 months, while the OECD includes those who experienced violence at any point in their lives Therefore, we would expect the OECD indicator to be larger than the UN indicator for most countries, given that many more women would have experience violence at some point in their lives.

Another limitation is that the UN indicator comes from many different reference years. Some countries have collected this data in more recent years (2015), while others last collected this data in the year 2000. This discrepancy makes comparison between countries and between indicators more difficult, since peoples’ attitudes towards domestic violence and countries’ laws and regulations may have changed significantly over the years, potentially reducing the prevalence of domestic violence. In addition, all the data from the OECD indicator was collected in 2019, much more recently than most of the UN data. Nonetheless, it would be interesting to find out whether both indicators are consistent with each other.

The figure above shows two maps of the world, with colored markers to represent the prevalence of violence in different countries. The first map visualizes the OECD indicator, while the second map visualizes the UN indicator. As expected, we can see that the OECD map has much darker and larger markers in general than the UN map, suggesting that the prevalence of violence is higher in the OECD data. This makes sense since the OECD measured the prevalence of violence over womens’ lifetimes, while the UN only measured violence in the past 12 months.

The two indicators also appear to be consistent across different regions. For instance, in both maps, Europe and Northern America appear to have less prevalence of domestic violence, as shown by the smaller, lighter markers. In contrast, markers across Africa, the Middle East and South America are consistently darker and larger, revealing patterns of higher prevalence of domestic violence.

[1] 0.7561751

Mapping Attitudes Towards Violence

Issues with the Dependent Variable Data and Limitations

For our dependent variables, we found some recurring data limitations.

Using Income Groups might reveal interesting patterns in what regards how differently countries with different income levels deal with violence against women, but it tells little about what are the causes of such discrepancies. Hence, whereas our graphs were helpful in revealing the path in which research can take, it did not necessary reveal to us what could be some of the causes for our indicators. Finally, using Income Group as a category might ignore nuances in regard to the socio-economic development of different countries. For example, countries like Brazil are classified as Upper Middle Income, but it also has one of the largest inequalities in the world (Gini Coefficient = 53.3), ranking 9th globally. Such incredible inequality could be driving some of our dependent variable, which separating countries by Income Group would not necessarily reveal.

One tidiness issue in the data is that the data set is in long format, which makes it difficult to visualize the relationship between the different indicators. Therefore, we had to transform the data set into wide format to create these plots.

There are also some discrepancies between what the indicators are trying to measure and what they actually measure. Quantifying such a large-scale phenomenon as violence against women is a non-trivial effort, since violence comes in many diverse forms. For example, the prevalence in lifetime indicator only quantifies violence from an intimate partner, excluding levels of harassment that come from outside intimate circles, such as sex trafficking. Similarly, to measure “attitude towards violence,” the data set creators used “the percentage of women who agree that a husband/partner is justified in beating his wife/partner under certain circumstances.” Why is this particular question used as a proxy for attitude towards violence? Why not ask whether non-intimate partners are also justified in beating women? Are there better or more comprehensive questions to gauge attitude towards violence?

In addition, for the “laws on domestic violence” indicator, the dataset creators do not explain how they quantified the abstract concept of “laws and practices.” They also do not specify what they mean by laws that “fully discriminate against women’s rights.” This makes it difficult to determine the accuracy of the indicator.

The dataset also does not contain all the countries in the world.

If we perform an anti-join of the countries dataset with the violence dataset, we can see that there are around 82 missing countries. Most of these countries are small islands that may not have enough data on violence against women.

Visualizing the Independent Variables

A Note on our Independent Variables Dataset

Our independent variables come from the Women’s Economic Opportunities Index, a composite index created by the Economist Intelligence Unit that evaluates whether a country’s environment is favorable towards female entrepreneurs and employees (“Women’s Economic Opportunity 2012” n.d.). The index was measured in two years: first in 2010, and later in 2012. We use the most recent 2012 dataset in our analysis. We downloaded the full dataset of this index as an Excel workbook from the Economist Intelligence Unit’s website. This original version can also be found in this repository under data/violence_factors. However, this Excel workbook contains several sheets and is not readable by dplyr’s csv importer, so we had to manually select the sheet we needed from Excel and export it in a .csv format to import it into R Studio. We used the Excel sheet named Data2012, which contained the 2012 index score for every country, in addition to the breakdown of each score into its categories and sub-categories. Since the sheet was not in csv format and contained several merged columns and rows, we had to clean it up manually in Excel before importing it as a csv file into R Studio. The cleaned version of the sheet can be found in the file data/violence_factors/woe_data_cleaned.csv.

Deciding to plot histograms of each of the four indicators under the WEO Index as well as the WEO Index its self we see that the education and legal status indicators are skewered more to the right of the scale closer to the 100 range which is more desirable. While for access to finance the values are more clustered around the center of the scale. The women opportunity index as well is fairly skewered to the right of the scale. The indicator labour policy is fairly clustered between the range of 25 to 75.

Limitations of the Independent Variables

One important limitation is the difference in dates. Whereas out dependent variable explores data collected in 2019, our independent variable (WEOI) is from 2012. It might be the case that countries have significantly improved or worsen their scores in the WEOI in the time between 2012 and 2019, which hence could be affecting our analysis.

The variables we have explored in the Women’s Economic Index Composite index are composite variables. Hence, even if we are looking for one specific variable that could particularly be causing higher levels of Attitude Towards Violence, this would not particularly be helpful. We would have to pay attention to the variables behind the larger indicators we are investigating. Besides that, being composite variables give large possibility of confounding variables to complicate our analysis. In other words, all the subcategories for each indicator are correlated, thus increasing the chances of having an unseen confounding variable.

Similar data limitations with our dependent variables return here, as well. The index has the very complicated way to measure an abstract concept such as “women’s economic opportunities.” Selecting which variables are put together to form one single indicator involves many human biases and interested/bias decisions. For example, since the indicator is developed by The Economist, many of the variables hint that they particularly considered important private sector entrepreneurship and initiatives as a measure for women’s opportunities. Same is applied to the importance they seem to attach in women’s entrance in finance and business. There could possibly been equally rigorous indexed that take into acount public policy and state-sponsored opportunities.

This is also connected with the fact that many of the qualitative indicators were created by the Economist Intelligence Unit rather than being collected by national statistics sources, which might further increase human and organizational bias in the collection and analysis of the data.

A final limitation is that our independent variable only includes 128 countries.

Part 2: Predicting Attitude Towards Violence

Correlation Coefficients

Having explored out dependent variables, we proceed to exploring the relationship between out dependent variables and our chosen independent variable – The Women’s Economic Opportunity Index. (WOEI)

todo Reem: fix correlation matrix

The matrix reveals that attitude towards women seem to negatively correlated with many of the indicators in the WEOI, especially education and training. However, it also evident that the indicators seem to be strongly correlated positively with each other as well. Such interactions could turn difficult to see confounding variables in our study.

           [,1]
attitude  -0.68
law       -0.17
prev_viol -0.37

Causal Diagram

From drawing correlations between the WEOI’s and the dependent variables, we notice that Attitude Towards Violence seems to offer the highest level of correlation (-0.67). For that reasons, we continue our analysis focusing on this variable in specific. Having chosen to focus on Attitude towards violence, we drew a causal diagram schematizing what can be the causes of higher levels of attitude towards violence.

When making our causal diagram we observed that the four indicators that are causing attitudes towards violence are: labour policy and practice, education and training, woman’s legal status and access to finance which fall under the Woman’s Economic Opportunity Index. We also observed a separate confounding variable affecting attitudes towards violence is gender segregation.

Regressions

Next, we plot and calculate the correlation between the Women’s opportunity index and Attitude Towards Violence further:

 Family: gaussian 
  Links: mu = identity; sigma = identity 
Formula: prev_viol ~ womens_opp_index 
   Data: relevant_factors (Number of observations: 89) 
  Draws: 4 chains, each with iter = 2000; warmup = 1000; thin = 1;
         total post-warmup draws = 4000

Population-Level Effects: 
                 Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS
Intercept           46.00      5.46    35.16    56.78 1.00     3576
womens_opp_index    -0.34      0.09    -0.51    -0.16 1.00     3628
                 Tail_ESS
Intercept            3090
womens_opp_index     2736

Family Specific Parameters: 
      Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS Tail_ESS
sigma    14.35      1.10    12.39    16.69 1.00     3207     2606

Draws were sampled using sampling(NUTS). For each parameter, Bulk_ESS
and Tail_ESS are effective sample size measures, and Rhat is the potential
scale reduction factor on split chains (at convergence, Rhat = 1).

When regressing the women’s oppurtunity index with attitudes towards violence we found that with evey one unit increase in woman’s oppurtunity index there is a negative decrease of -0.39 in attitudes towards violence with a confidence interval of -0.56 and -0.22 and a uncertainty of 0.09.

 Family: gaussian 
  Links: mu = identity; sigma = identity 
Formula: attitude ~ labour_policy 
   Data: relevant_factors (Number of observations: 89) 
  Draws: 4 chains, each with iter = 2000; warmup = 1000; thin = 1;
         total post-warmup draws = 4000

Population-Level Effects: 
              Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS
Intercept        48.09      6.27    36.02    60.68 1.00     3762
labour_policy    -0.50      0.11    -0.72    -0.29 1.00     3788
              Tail_ESS
Intercept         2779
labour_policy     3035

Family Specific Parameters: 
      Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS Tail_ESS
sigma    16.56      1.25    14.32    19.23 1.00     3853     2777

Draws were sampled using sampling(NUTS). For each parameter, Bulk_ESS
and Tail_ESS are effective sample size measures, and Rhat is the potential
scale reduction factor on split chains (at convergence, Rhat = 1).

The estimate for Labour Policy and Practice is -0.47. Hence, With every one unit increase in Labour Policy and Practice there is a decrease of -0.47 in attitudes towards violence. The confidence levels range from -0.62 and -0.31, with an uncertainty of 0.08.

 Family: gaussian 
  Links: mu = identity; sigma = identity 
Formula: attitude ~ finance_access 
   Data: relevant_factors (Number of observations: 89) 
  Draws: 4 chains, each with iter = 2000; warmup = 1000; thin = 1;
         total post-warmup draws = 4000

Population-Level Effects: 
               Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS
Intercept         44.15      3.54    37.26    51.20 1.00     3848
finance_access    -0.50      0.07    -0.63    -0.37 1.00     3818
               Tail_ESS
Intercept          2666
finance_access     2677

Family Specific Parameters: 
      Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS Tail_ESS
sigma    14.56      1.13    12.56    16.95 1.00     4047     2970

Draws were sampled using sampling(NUTS). For each parameter, Bulk_ESS
and Tail_ESS are effective sample size measures, and Rhat is the potential
scale reduction factor on split chains (at convergence, Rhat = 1).

The estimate for Access to Finance is also -0.47. Hence, With every one unit increase in Access to Finance there is a decrease of -0.47 in attitudes towards violence. The confidence levels range are a bit smaller then from Labour Policy and Practice, ranging from -0.57 and -0.37, with an also smaller uncertainty of 0.05.

 Family: gaussian 
  Links: mu = identity; sigma = identity 
Formula: attitude ~ education 
   Data: relevant_factors (Number of observations: 89) 
  Draws: 4 chains, each with iter = 2000; warmup = 1000; thin = 1;
         total post-warmup draws = 4000

Population-Level Effects: 
          Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS Tail_ESS
Intercept    59.10      3.99    51.46    66.97 1.00     3800     2869
education    -0.62      0.06    -0.74    -0.50 1.00     3804     2710

Family Specific Parameters: 
      Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS Tail_ESS
sigma    12.53      0.97    10.80    14.60 1.00     4336     3089

Draws were sampled using sampling(NUTS). For each parameter, Bulk_ESS
and Tail_ESS are effective sample size measures, and Rhat is the potential
scale reduction factor on split chains (at convergence, Rhat = 1).

The estimate for Education and Training is -0.59. Hence, With every one unit increase in Education and Training there is a decrease of -0.59 in attitudes towards violence. This is a more significant change then the previous two variables explored.The confidence levels range from -0.69 and -0.49, with an uncertainty of 0.05.

 Family: gaussian 
  Links: mu = identity; sigma = identity 
Formula: attitude ~ legal_status 
   Data: relevant_factors (Number of observations: 89) 
  Draws: 4 chains, each with iter = 2000; warmup = 1000; thin = 1;
         total post-warmup draws = 4000

Population-Level Effects: 
             Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS
Intercept       80.19      6.69    67.34    93.12 1.00     3389
legal_status    -0.83      0.09    -1.01    -0.66 1.00     3448
             Tail_ESS
Intercept        2780
legal_status     3002

Family Specific Parameters: 
      Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS Tail_ESS
sigma    13.24      1.01    11.48    15.40 1.00     3483     2694

Draws were sampled using sampling(NUTS). For each parameter, Bulk_ESS
and Tail_ESS are effective sample size measures, and Rhat is the potential
scale reduction factor on split chains (at convergence, Rhat = 1).

The estimate for Women’s Legal Status is -0.80. This is the highest estimate we found. Hence, With every one unit increase in Women’s Legal Status there is a decrease of -0.80 in attitudes towards violence. The confidence levels range from -0.95 and -0.66, with an uncertainty of 0.07. This is the indicator with strongest correlation with a reduction in Attitude Towards Violence.

Also noticing that the confidence intervals of each of the regressions is fairly small which means the uncertainty interval is fairly small. Additionally, the standard error of each coefficient were low.

With those results in mind, we draw a multivariate regression:

 Family: gaussian 
  Links: mu = identity; sigma = identity 
Formula: attitude ~ labour_policy + education + legal_status + finance_access 
   Data: relevant_factors (Number of observations: 89) 
  Draws: 4 chains, each with iter = 2000; warmup = 1000; thin = 1;
         total post-warmup draws = 4000

Population-Level Effects: 
               Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS
Intercept         69.77      6.73    56.37    83.04 1.00     4291
labour_policy      0.21      0.11    -0.02     0.43 1.00     4562
education         -0.42      0.13    -0.67    -0.18 1.00     3457
legal_status      -0.45      0.16    -0.75    -0.14 1.00     4496
finance_access    -0.04      0.10    -0.23     0.15 1.00     3783
               Tail_ESS
Intercept          3272
labour_policy      2718
education          2304
legal_status       2738
finance_access     2609

Family Specific Parameters: 
      Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS Tail_ESS
sigma    12.10      0.96    10.40    14.17 1.00     4784     2673

Draws were sampled using sampling(NUTS). For each parameter, Bulk_ESS
and Tail_ESS are effective sample size measures, and Rhat is the potential
scale reduction factor on split chains (at convergence, Rhat = 1).

When running a multivariate regression of all of the four indicators with attitude towards violence we found that the indicators education, legal status and access to finance still hold a negative decrease effect on attitudes towards violence with every one unit increase, however labour policy now has a positive increase in attitudes towards violence with every one unit increase which is interesting to see. The confidence intervals of each has increased as well as the standard error coefficient which means there is a larger uncertainty range.

Even when running the multivariate regression the best indicator remains still to be the legal status with the highest negitave decrease in attitudes towards violence with the coefficients of -0.45. Therefore, further exploration on the indicator should be done.

Breakdown of Legal Status

 Family: gaussian 
  Links: mu = identity; sigma = identity 
Formula: attitude ~ addressing_violence + citizen_rights + prop_ownership + adol_fertility + contraceptive_use + cedaw + political_part 
   Data: legal_status_data (Number of observations: 89) 
  Draws: 4 chains, each with iter = 2000; warmup = 1000; thin = 1;
         total post-warmup draws = 4000

Population-Level Effects: 
                    Estimate Est.Error l-95% CI u-95% CI Rhat
Intercept              77.76     10.69    56.17    98.86 1.00
addressing_violence    -0.06      0.08    -0.21     0.09 1.00
citizen_rights         -0.13      0.13    -0.39     0.14 1.00
prop_ownership         -0.19      0.06    -0.31    -0.08 1.00
adol_fertility         -0.16      0.07    -0.30    -0.02 1.00
contraceptive_use      -0.18      0.07    -0.32    -0.05 1.00
cedaw                  -0.09      0.06    -0.21     0.03 1.00
political_part          0.08      0.08    -0.07     0.24 1.00
                    Bulk_ESS Tail_ESS
Intercept               4872     3535
addressing_violence     3926     2857
citizen_rights          4068     3060
prop_ownership          3726     2768
adol_fertility          3266     2630
contraceptive_use       3445     2691
cedaw                   4287     3084
political_part          4427     3107

Family Specific Parameters: 
      Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS Tail_ESS
sigma    12.72      0.99    11.01    14.80 1.00     3251     2932

Draws were sampled using sampling(NUTS). For each parameter, Bulk_ESS
and Tail_ESS are effective sample size measures, and Rhat is the potential
scale reduction factor on split chains (at convergence, Rhat = 1).

When running a regression on further exploration of the indicator legal status, we found that within the factors under legal status the one with the highest effect on attitudes towards violence was property ownership with a negative decrease of -0.19 with every on unit increase with a uncertainty of 0.06 and a confidence interval of -0.31 and -0.07.

As seen from the graph as well as from the regression we ran on the indicator legal status, a higher level on property ownership rights results in a decrease in positive attitudes towards violence. While it is possible for outliers to be impacting the data, the graph demonstrates a linear decrease pattern between the two.

We then continued by observing the relationship between contraception usage, which is a factor that falls under the legal status indicator, and attitudes towards violence. That’s because in our previous analysis contraception usage came second highest in negatively decreasing attitudes towards violence with a decrease of -0.18 with a confidence interval of -0.32 and -0.04 with an uncertainty of 0.07.

From the graph we see that there is a negative relationship between contraception use and attitudes towards violence. The linear decrease pattern is more observable in this graph as the data is fairly clustered around the line as well as a decrease motion is visible while there still is a few outliers.

Upon reflecting on what are some of the factors that might be influencing or causing women’s legal status, we moved our research forward by analyzing the role that democracy has on women’s legal status.

Exploring the Effects of Democracy

 Family: gaussian 
  Links: mu = identity; sigma = identity 
Formula: legal_status ~ democracy_index 
   Data: relevant_factors (Number of observations: 89) 
  Draws: 4 chains, each with iter = 2000; warmup = 1000; thin = 1;
         total post-warmup draws = 4000

Population-Level Effects: 
                Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS
Intercept          34.07      3.51    27.29    40.93 1.00     4137
democracy_index     6.06      0.53     5.01     7.08 1.00     4042
                Tail_ESS
Intercept           2779
democracy_index     2925

Family Specific Parameters: 
      Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS Tail_ESS
sigma     9.95      0.76     8.57    11.59 1.00     3777     3003

Draws were sampled using sampling(NUTS). For each parameter, Bulk_ESS
and Tail_ESS are effective sample size measures, and Rhat is the potential
scale reduction factor on split chains (at convergence, Rhat = 1).

 Family: gaussian 
  Links: mu = identity; sigma = identity 
Formula: attitude ~ democracy_index 
   Data: relevant_factors (Number of observations: 128) 
  Draws: 4 chains, each with iter = 2000; warmup = 1000; thin = 1;
         total post-warmup draws = 4000

Population-Level Effects: 
                Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS
Intercept          66.63      4.45    57.99    75.71 1.00     3939
democracy_index    -6.98      0.73    -8.46    -5.58 1.00     4184
                Tail_ESS
Intercept           3007
democracy_index     2876

Family Specific Parameters: 
      Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS Tail_ESS
sigma    17.74      1.15    15.69    20.19 1.00     3932     3086

Draws were sampled using sampling(NUTS). For each parameter, Bulk_ESS
and Tail_ESS are effective sample size measures, and Rhat is the potential
scale reduction factor on split chains (at convergence, Rhat = 1).

 Family: gaussian 
  Links: mu = identity; sigma = identity 
Formula: attitude ~ democracy_index + legal_status 
   Data: relevant_factors (Number of observations: 89) 
  Draws: 4 chains, each with iter = 2000; warmup = 1000; thin = 1;
         total post-warmup draws = 4000

Population-Level Effects: 
                Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS
Intercept          78.96      6.80    65.76    92.58 1.00     4065
democracy_index    -1.49      1.11    -3.64     0.79 1.00     2429
legal_status       -0.69      0.14    -0.96    -0.41 1.00     2521
                Tail_ESS
Intercept           3184
democracy_index     2439
legal_status        2817

Family Specific Parameters: 
      Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS Tail_ESS
sigma    13.15      0.98    11.36    15.22 1.00     3385     2655

Draws were sampled using sampling(NUTS). For each parameter, Bulk_ESS
and Tail_ESS are effective sample size measures, and Rhat is the potential
scale reduction factor on split chains (at convergence, Rhat = 1).

Analysis and Interpretation of Results

When we ran the regression between attitudes towards violence individually with each of the four indicators from the WEOI, we noticed that women’s legal status seemed to be the most strongly negatively correlated with the dependent variable (-0.80). Running a multivariate regression with all indicators, however, decreased the correlation of women’s legal status to (-0.45). Even though it remains moderately correlated, such interaction reveal that it is difficult to isolate one single cause for the increase or decrease of levels in attitudes towards violence. Interestingly, when we ran the multivariate regression, the indicator Labour Policy and Practice went from a negative coefficient of -0.47 to 0.21.

A counterfactual that arises from our research is imagining a country in which the women’s legal status was significantly lower than the rest of the indicators. Would such a country have lower levels of attitude against violence (less women agreeing that violence is justified)? Even though we cannot find such a case within the indicators of the WOEI, the exploration on the dependent variables highlighted the Russian Federation as having laws on domestic violence that completely discriminated against women but had decent levels in prevalence of violence in lifetime and attitude towards women. A further study on the case of the Russian Federation could reveal the effectiveness of women’s legal status (and specifically laws against domestic violence) in causing a decrease in violence against women.

we found that among the sub-indicators that form Women’s Legal Status, property rights and contraceptive use were among the ones driving Attitude Towards Violence the most, with respectively coefficients of -0.19 (uncertainty of 0.06) and -0.18 (uncertainty of 0.07).

When analyzing the causal relationship between the democracy index and the legal status of women, we notice that democracy seems to be driving Women’s legal status up, which in turn drives attitude towards violence to decrease. This was an interesting finding: the higher the levels of democracy, the higher the status of women legally, which in turn produce lower levels of acceptance of violence against women.

Hence, our findings do corroborated that strengthening democracy and, hence, the legal status of women might be a path for decreasing attitude towards violence. Increasing women’s legal status can be also be done by significantly increasing women’s property rights and contraception use.

Obviously, it might be far-fetched to suggest democracy as a remedy for decreasing attitude towards violence across the globe, as it suggests this is applicable to all cultures. It is important to take into account different political and socio-economy, as well as historical contingencies, when designing public policy and in the process of policy-making.

Conclusions

We cannot single out or highlight one single indicator in the WEOI that is most influential in decreasing violence against women. Our research highlights the complexities and the varying number of factors that result in the persistence of violence against women. However, it still might be useful to investigate further the sub-indicators within each indicator for the WEOI and analyze what they can provide as insights. Singling out countries, like Russia, and analyze closely what is the situation in violence against women through both quantitative and qualitative research can further reveal indicators that can be useful in measuring degrees of such forms of gendered violence.

While the four indicators we observed and analysed did in fact demonstrate a decrease in attitudes towards violence there are confounding variables that we perhaps have not considered which makes it hard to come to a causal conclusion. It is also interesting to see that indicators when observed separately demonstrate an effect on attitudes towards violence however when observed alongside other indicators have completely opposing effects.

This studies shows how data is complex and immense while certain indicators and factors might demonstrate an effect on a variable the reason for such an effect could be due to many confounding variables and other reasoning to what one might original have thought.

With that being said, the “top-to-bottom” structure of this research, in which we started from one single index and proceeded to slowly peeling off its layers was extremely insightful in terms of shedding light into how complex indexes and composite variables are constructed. It highlights how several measures, some obvious, some more ominous, converse to each other to finally measure an index. It also reminds us that the construction of such indexes are attempts to measure incredibly abstract concepts into truly useful ones. Whereas this was helpful for us in schematizing causal diagrams, we also found ourselves constantly puzzled on what we were truly measuring.

References